47 results about "Nonlinear classification" patented technology

The invention discloses a super-resolution face recognition method based on deep belief networks, and relates to the technical field of face recognition. From a cognitive perspective, it is believed that an intrinsic relation exists between mutually corresponding face images differing in resolutions. But previous studies show that the method for expressing the intrinsic relation by linear approximation is restricted by linear approximation. Therefore, it is believed that the intrinsic relation is nonlinear. In view of outstanding performances of an artificial neural network on the nonlinear classification problem, a neural network algorithm is adopted to capture the nonlinear relation of the mutually corresponding face images differing in resolutions under the condition of posture change. Both theoretical research and neurophysiological research show that a deep structure, such as a system constructed by multiple layers of nonlinear processing units, should be constructed to build an intelligent processing system. According to the face recognition method, deep belief networks are adopted to extract a common nonlinear structure shared by mutually corresponding face images differing in resolutions.

The invention discloses an SVM (support vector machine)-based nonlinear damage removing device of a coherent optical communication system. The device disclosed by the invention comprises a chromatic dispersion compensating unit, an SVM array unit, an SVM training unit, a logical processing unit and a sign deciding unit, wherein the chromatic dispersion compensating unit is used for compensating optical fiber dispersion applied to receiving signals; the SVM array unit is used for carrying out binary classification on the receiving signs according to different classification rules via a plurality of SVMs; the SVM training unit is used for defining a classification hyper plane of each SVM in the device according to a certain training sequence; the logical processing unit is used for performing logical operation on classification results of each SVM to acquire type signs corresponding to the signals; and the sign deciding unit is used for decoding the signal type signs to binary sequences corresponding to the signals. The SVM-based nonlinear damage removing device of the coherent optical communication system can be used for removing optical nonlinear damage to the signals by virtue of the nonlinear classification characteristic of the SVM. Without requirements of knowing the nonlinear characteristic of optical fibers, the nonlinear damage removing effect is ensured and the processing complexity is only decided by the number of a small quantity of support vectors.

The invention discloses a method for distinguishing fog concentration in intelligent image monitoring of a power transmission line, which adopts a multi-feature synthesis method and comprises the following steps of: extracting a plurality of features related to fog in a picture, and synthesizing the plurality of features extracted from each image into a feature parameter table to form a two-dimensional feature map; utilizing a deep learning convolutional neural network to train and model the characteristic parameter graphs of a large number of sample images; and judging whether the unknown image has fog or not by using the trained model, and distinguishing the concentration levels of the fog, namely clear fog, medium fog and dense fog. According to the method, a plurality of deep and irrelevant features of the provider are integrated to serve as image feature parameters, and then a nonlinear classification algorithm model is trained for a large number of sample images by using the convolutional neural network, so that the problem that the accuracy of detecting whether the image is foggy is low is solved.

The invention relates to an optimal classification electric vehicle charging state monitoring method comprising the following steps: respectively obtaining vehicle-end real time BMS data in a chargingprocess and charging pile end real time BMS data in the charging process; storing the obtained data in a charging management platform; using the charging management platform stored data to build a high dimension data model, and carrying out non-linear classification; comparing the non-linear classification result with a real fault result so as to obtain a relation between the classification and the fault; allowing a system manager to increase communication protocol message parameters used by the charging pile and BMS in the charging process according to non-linear classification result or therelation between the classification and the fault. The monitoring method can acquire the real time data of the charging pile end, can acquire the vehicle-end real time data, can use a feature extraction method based on the non-linear optimal classification to analyze the data, and can real time monitor the health states of the power battery and the charging pile in the charging process, thus providing basis for diagnosing the power battery performance of the electric vehicle.

The invention discloses a load identification method based on a support vector machine and V-I curve characteristics, relates to the technical field of load identification systems, and overcomes the defect of possible false identification in non-intrusive load identification by using a combination of a V-I curve and harmonic waves as a load mark. A plurality of load marks are formed by using shapecharacteristics of a V-I curve track, the correct rate of load identification is increased, and the defect that a V-I curve is difficult to identify a small load is overcome by being assisted by harmonic characteristic identification obtained by Fourier transform of an electrical quantity; the nonlinear classification problem is better processed through SVM-based load identification, and dimensionality disasters cannot be caused; according to the method, machine learning of small samples can be processed, the problems of local minimum, over-learning and under-learning are avoided, the load identification result is more accurate, and the identification degree is improved. In addition, the method takes a non-intrusive mode as a starting point, and has the characteristics of economy, practicability and easiness in implementation.

The invention discloses a rapid network representation learning algorithm based on a width learning system. The method comprises the following steps of S1, importing a network graph module based on atext, parsing and storing a network topological structure in a dictionary format, wherein keys in the dictionary represent network nodes, values corresponding to the keys form a list and represent a node sequence at the other end of the edge where the nodes are located; S2, performing random walk on the network nodes to generate a walk sequence; S3, constructing a network representation learning model based on a width learning system, taking the walking sequence generated in the step S2 and a representation vector with the dimension of K as input, generating a feature vector of a network nodein a feature vector layer, and enhancing the nonlinear classification capability of a network representation learning model by introducing an activation function in an enhancement vector layer to finally realize text-based network multi-label classification. A width learning system model is adopted in the algorithm, and representation learning of network nodes can be rapidly completed.

The invention provides a remote sensing rainfall error correction method and system based on nonlinear classification regression analysis. The remote sensing rainfall error correction method comprises the following steps: establishing a target drainage basin rainfall geographic space-time information database; dividing a target drainage basin rainfall station, determining a training set and a test set, and calculating an initial error correction field at the target drainage basin rainfall station; determining a correction domain by taking observation field geographic information corresponding to the training set as a benchmark, and constructing a nonlinear classification regression model set in the correction domain on the basis of a support vector machine classification regression theory period by period; automatically screening a correction domain and nonlinear classification regression model parameters according to the geographic information of the test set, estimating a rainfall error field, and performing error correction and precision evaluation on a background field of the test set; utilizing a nonlinear classification regression model to construct rainfall error fields with different spatial resolutions to realize downscaling processing, and carrying out grid-by-grid and period-by-period error correction on a remote sensing rainfall product. According to the embodiment of the invention, the remote sensing rainfall data set with higher precision and higher resolution can be generated as required.

The invention discloses a myoelectricity recognition method based on a two-parameter kernel optimization type extreme learning machine. The method comprises the following steps: firstly, extracting four paths of electromyographic signals and extracting corresponding average amplitude, variance, Wilson amplitude and wavelet energy coefficients, then fusing the characteristics, and finally, transmitting the fused characteristics to a dual-parameter optimization type extreme learning machine. According to the dual-parameter optimization type extreme learning machine, on the basis of the extreme learning machine, a Gaussian kernel function is introduced, all parameters are set and optimized by minimizing an output weight matrix, a neural network structure is constructed, and the problem of minimizing an output error of the extreme learning machine is converted into the problem of minimizing an output weight. Compared with a traditional extreme learning machine, the method has the advantages that the function approximation capability is stronger, the nonlinear classification processing capability is stronger, and compared with other common classifier algorithms, the method also has higher accuracy and shorter operation time.

The invention provides an SVM (Support Vector Machine)-based Web partitioning method, which comprises the steps of: partitioning all Websites into N groups; taking K=1, 2, 3, ......,N, for the value of each K, selecting Website samples in the 1st-(K-1)th, (K+1)th-Nth groups, and initializing LibSvm training; training LibSvm; storing a trained SVM model; selecting a Website sample in a Kth group to perform a Web partitioning test; and saving the Web partitioning test result. According to the method provided by the invention, the generalization capability of the SVM is strong, fault tolerance and classification can be better performed when louder noise data are processed. The accuracy rate of coordinates established by a network coordinate system is approximately 80 percent, the problem of nonlinear classification can be solved by the SVM, the number of classification by the SVM is fixed, the extreme condition that no crawler crawls on a website is avoided, and the uncertainty of partitioning the number of a set in the clustering algorithm is overcome by using the classification algorithm.

In order to improve the prediction accuracy of rainfall type landslide and overcome the problem that an existing prediction method is difficult to process a multi-factor nonlinear relation, the invention discloses a modeling method for predicting multi-factor induced landslide based on deep learning, a traditional DBN algorithm is improved, a momentum learning rate, Dropout and Softmax technologies are introduced, convergence difficulty or local optimum is avoided. The problem of overfitting is reduced, and nonlinear classification optimization and prediction of multiple influence factors causing rainfall-type landslide are realized. A simulation experiment result verifies the accuracy of the model provided by the invention, and a new thought is provided for rainfall type landslide prediction by using a deep learning method.

An ultrasonic interferometer for the characterization of the matter (solids, fluids & gases) in a medium is described, along with a method of using the same. The interferometer accurately determines the travel time of the multi-frequency ultrasonic signal in the matter that is being queried. By carefully selecting the design of the multi-frequency ultrasound signal, various properties of the material can be derived using a trainable classification system to classify or recognise the substance, or state of a process. The apparatus exploits the normally undesirable higher harmonics characteristics of the piezoceramic transducer to gain penetration through a spatial-frequency window that is not suitable for the higher frequency signals that are required to achieve the measurements.

The invention discloses a short-term electric power load prediction modeling method, comprising the steps that based on the feature of electric load and by means of core major constituent analysis, non-linear feature extraction is conducted on the affecting factors influencing the electric load; redundancy and co-linear information among variables can be eliminated, and the principal component feature components affecting the electric load can be obtained; the electric load can be predicted by means of BP nerve network having strong non-linear analysis capability. The invention is advantageousin that under the precondition that more information of the original data is fully reserved, the core major constituent analysis effectively conducts feature dimension reduction on the electric loaddata, and thereby the structure of the BP nerve network is optimized, and prediction speed and precision can be greatly improved.

To increase the timeliness, objectivity, and efficiency in evaluating surgical procedures such as those performed by ophthalmology residents' learning of cataract surgery, an automatic analysis system for surgeries such as cataract surgery is provided to assess performance, particularly in the capsulorrhexis step on the Kitaro simulator. Computer vision technologies are employed to measure performance of this critical step including duration, centrality, circularity, size, as well as motion stability during the capsulorrhexis procedure. Consequently, a grading mechanism is established based on either linear regression or non-linear classification via Support Vector Machine (SVM) of those computed measures. Comparisons of expert graders to the computer vision based approach have demonstrated the accuracy and consistency of the computerized technique.

This application provides methods to differentiate between subjects with autism spectrum disorder (ASD) and typically developing (TD) or developmentally delayed (DD) subjects using miRNA and/or microbiome levels detected in saliva samples and patient information. The method can be used to monitor the progress of ASD and guide its treatment. RNA-seq, qPCR, or other methods determine counts and abundance of miRNA or microbiomes. MicroRNA and/or microbiome sequencing data are refined by normalization to expression levels or abundance of time-invariant miRNAs and/or microbial RN As to control for time of sample collection or to compensate for circadian fluctuations in these levels. Multivariate logistic regression and nonlinear classification techniques are further used to select a panel of miRNAs and microbiomes that accurately differentiate between subjects with ASD, DD, and TD in subjects with an unknown ASD status. These panels of miRNAs and microbiomes may be developed into a RNA assay kit.

The invention discloses an environment pollutant source analysis method based on corresponding analysis and multiple linear regression, and the method comprises the following steps: firstly, based on pollution source sample data, employing a corresponding analysis method to recognize a pollution source, and determining the number of main factors; secondly, calculating the contribution rate of the pollution source of the factor load by utilizing multiple linear regression, and realizing source analysis of the characteristic pollutants. According to the pollution source identification method based on corresponding analysis and multiple linear regression, the pollution source is identified by using a corresponding analysis method, the contribution rate of the pollution source is calculated by compounding a multiple linear regression method, and a factor load identification process is regarded as a nonlinear classification process which is a multi-factor comprehensive classification problem. The method is a mode recognition process, is high in practicability, has wide popularization and application values, and provides reliable technical guarantee for an environment management department to deal with pollution accidents and control pollution risks.

The invention provides a hierarchical state estimation method based on model prediction extended Kalman filtering. The method comprises the steps of: firstly, converting a nonlinear coupling system model into a linear classification model and a nonlinear classification model, performing prediction based on the linear classification model and the nonlinear classification model on a state estimationvalue at the previous moment through first-stage extended Kalman filtering, and obtaining a linear state component estimation value; secondly, using the linear state component estimation value for feeding back the nonlinear grading model, and obtaining a nonlinear prediction model; and finally, performing nonlinear prediction model-based prediction on the linear state component estimated value byusing secondary extended Kalman filtering to obtain a nonlinear state component estimated value. According to the hierarchical state estimation method provided by the invention, the system state estimation precision is improved, the calculation dimension of the estimation process is reduced, and the hierarchical state estimation method is suitable for high-dimensional and linear state-decoupled coupled complex system state estimation.

The invention relates to an image classification method based on random Fourier feature transformation, and belongs to the technical field of image classification. The classification method comprises the steps of 1 preprocessing a training image, obtaining a preprocessed image, wherein preprocessing comprises graying, geometric transformation, image enhancement, image segmentation and image denoising; 2, performing feature extraction on the preprocessed image to obtain image features, and constructing a training set, wherein the image features comprise color features, texture features, algebraic features and transformation features; 3, training similar image features to obtain a new weight vector and a separation distance; and 4, performing preprocessing, feature extraction, random Fourier transform and classification on the to-be-classified image to obtain a classification result. According to the method, the accuracy of small sample and single sample images is high; the time and space complexity is low; the problems of neural network result selection and local minimum value are avoided; and the generalization of high-dimensional and nonlinear classification problems is good.

The invention relates to a classification method based on channel measurement, and belongs to the technical field of channel identification environment backscatter communication and machine learning. The system comprises a transmitter, a receiver and a small radio frequency transceiver. The transmitter sends PN codes and sine waves; the receiver receives the signal reflected and absorbed by the small radio frequency transceiver; the method comprises the following steps: step 1, a receiver receives a signal sent by a transmitter; 2, performing feature extraction on the received signals to obtain a feature set X and a feature test set Xt respectively; step 3, training by using the feature set X to obtain trained classification model parameters; and step 4, classifying the feature test set Xt. According to the method, the time complexity and the space complexity are low, and high recognition accuracy can still be guaranteed under the condition of small sample data; the method has good generalization for high-dimensional and nonlinear classification problems, and can be used for feature extraction of channel receiving signals in most application scenarios.